Heat control parameter dynamic optimization method and system for liquid cooling lithium iron phosphate battery

Abstract

The invention discloses a heat management and control parameter dynamic optimization method and system for a liquid-cooled lithium iron phosphate battery, and relates to the field of battery heat management and control parameter dynamic optimization, and the method comprises the steps: building a carrier heat state offline training database based on a carrier state, a heat state, a cooling system and environment parameters, and carrying out the data cleaning and alignment; based on a LightGBM model, constructing a general thermal state prediction basic model of the carrier; verifying whether the thermal state of the predicted carrier is reliable or not by combining an actual monitoring result; dynamically optimizing the carrier general thermal state prediction basic model based on an online gradient descent algorithm; dynamically optimizing the flow velocity of the cooling liquid based on a heat balance equation; a sensor set is arranged and used for collecting current state parameters and thermal state actual data in real time. The method has the advantages that data driving and a physical model are fused, the heat management and control parameters are dynamically optimized through online learning, and the heat management precision, adaptability and safety of the liquid-cooled lithium iron phosphate battery are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery thermal management parameter dynamic optimization, in particular to a thermal management parameter dynamic optimization method and system for liquid-cooled lithium iron phosphate batteries. BACKGROUND

[0002] With the rapid development of new energy vehicle industry, the thermal safety problem of power batteries has become increasingly prominent. Lithium iron phosphate batteries have become one of the mainstream choices due to their high safety and long cycle life, but their thermal management work still faces severe challenges. A large amount of heat is generated during the charging and discharging process of the battery, and if the heat is not dissipated in time, it will cause the temperature of the battery pack to rise, the service life to decay, and even cause safety accidents such as thermal runaway. The liquid cooling system has become the mainstream thermal management solution for large-capacity battery packs due to its efficient and uniform cooling performance. The core is to dynamically adjust the cooling liquid flow rate and other parameters according to the thermal state of the battery to achieve precise temperature control.

[0003] Traditional thermal management strategies are usually based on preset thresholds or simple PID controllers, relying on a large number of bench tests and engineer experience, which is a static and passive control method. However, the actual operating conditions of vehicles are complex and variable, including different environmental temperatures, driving habits, road conditions, etc. These factors will dynamically affect the heat generation and dissipation balance of the battery. Static strategies cannot adapt to such dynamic changes, often showing control lag or over-regulation, resulting in low energy efficiency and insufficient temperature control accuracy. Therefore, developing an intelligent thermal management method that can real-time perceive state, predict trends and dynamically optimize parameters is of great significance for improving battery performance, ensuring safety and prolonging life, and is also an urgent need for current industry technology development.

[0004] The existing prediction method based on neural network and other machine learning large models is prone to rely on the scale and comprehensiveness of training data. For carriers with low usage frequency or special behavior patterns, it is difficult to obtain sufficient sample data for training, resulting in a significant decrease in model prediction accuracy and insufficient generalization ability, which cannot effectively capture diversified driving habits. Secondly, there is a "black box" effect and lack of physical constraints, and the prediction process lacks explainability. More critically, the prediction results may violate basic physical laws (such as energy conservation and thermodynamic principles), leading to a discrepancy between the output and the actual situation. This makes it difficult for existing technologies to effectively integrate data-driven learning ability with the explainability and safety of physical models, ultimately compromising between prediction accuracy and system reliability SUMMARY

[0005] To solve the above technical problems, a heat management parameter dynamic optimization method and system for a liquid-cooled lithium iron phosphate battery are provided. The technical solution solves the problem that the model prediction accuracy significantly decreases and the generalization ability is insufficient due to the difficulty in obtaining sufficient sample data for training for carriers with low usage frequency or special behavior patterns, which cannot effectively capture diversified driving habits. In addition, the "black box" effect and lack of physical constraints exist, the prediction process lacks explainability, and more critically, the prediction results may violate basic physical laws (such as energy conservation and thermodynamic laws), leading to a discrepancy between the output and the actual situation. This makes it difficult for the existing technology to effectively integrate data-driven learning ability with the explainability and safety of physical models, ultimately balancing the prediction accuracy and system reliability.

[0006] To achieve the above purposes, the technical solution adopted by the present application is:

[0007] A heat management parameter dynamic optimization method for a liquid-cooled lithium iron phosphate battery, comprising:

[0008] Based on the carrier state, thermal state, cooling system, and environmental parameters, a carrier thermal state offline training database is constructed, and data cleaning and alignment are performed;

[0009] According to the carrier thermal state offline training database, a carrier general thermal state prediction base model is constructed based on the LightGBM model, which is used to predict the carrier thermal state;

[0010] According to the carrier general thermal state prediction base model, combined with the actual monitoring results, verify whether the predicted carrier thermal state is reliable;

[0011] Based on the online gradient descent algorithm, combined with the driving data of the carrier, dynamically optimize the carrier general thermal state prediction base model to automatically learn driving habits;

[0012] According to the output results of the carrier general thermal state prediction base model, based on the heat balance equation, dynamically optimize the cooling liquid flow rate;

[0013] Set up a sensor group to real-time collect current state parameters and future time thermal state actual data for verifying prediction accuracy required by the model.

[0014] Preferably, the carrier general thermal state prediction base model is constructed based on the LightGBM model according to the carrier thermal state offline training database, which is used to predict the carrier thermal state, specifically comprising:

[0015] The carrier state parameters, cooling system parameters, and environmental parameters are used as input quantities, and the carrier thermal state parameters in the future period of time are used as output quantities;

[0016] The environmental parameter is used to represent the dynamic constraint on the cooling liquid inlet temperature and the system heat dissipation boundary, so as to improve the prediction accuracy of the model under different working conditions.

[0017] Based on the offline training database of the carrier thermal state, the database is divided into a training set, a verification set and a test set in the ratio of 7:2:1;

[0018] Based on the LightGBM framework, a gradient boosting decision tree algorithm is used to build a general thermal state prediction model for the carrier, which is composed of multiple regression trees.

[0019] Each regression tree splits nodes to minimize the weighted sum of squared errors. When splitting, the optimal feature and its split threshold are selected by maximizing the gain.

[0020] The initial learning rate is set to A. During the training process, the loss on the validation set is continuously monitored. If the loss does not decrease for 5 consecutive iterations, the current learning rate is halved

[0021] A custom loss function with safety penalty is constructed based on the mean square error formula to minimize the average temperature deviation of the battery.

[0022] The gradient boosting algorithm is used to optimize the model parameters. In each iteration, the first and second order gradients of the current model with respect to the custom loss function are calculated, and a new regression tree is added to fit the negative gradient direction, thereby reducing the loss.

[0023] When the absolute value of the change of the validation set loss function value in K consecutive iterations is less than the preset threshold a, it is determined that the model has converged, and the training process is terminated in advance to avoid overfitting.

[0024] The gain expression is:

[0025]

[0026] In the formula, Gain is the gain value, which is used to evaluate the pros and cons of the split point, G L , G R are the sums of the first order gradients of all samples in the left and right child nodes after splitting, H L , H R are the sums of the second order gradients of all samples in the left and right child nodes after splitting, λ is the L2 regularization coefficient, and γ is the minimum gain threshold required for splitting. Splitting only occurs when Gain>γ.

[0027] The expression of the custom loss function with safety penalty is:

[0028]

[0029] wherein, L is a custom loss value with safety penalty, M is the number of samples participating in training, Y true.i is the real value of the average battery temperature of the i-th sample, Y pred.i is the predicted value of the average battery temperature of the i-th sample, Y safe is a preset safety temperature threshold, and a is a safety penalty factor.

[0030] Preferably, the step of verifying whether the predicted thermal state of the carrier is reliable according to the carrier universal thermal state prediction base model in combination with the actual monitoring result specifically comprises:

[0031] According to the trained carrier universal thermal state prediction base model, based on the actual monitoring data of the carrier state parameters, cooling system parameters and environmental parameters, the carrier thermal state parameter data in a future period of time is obtained;

[0032] The thermal state parameter prediction data of the model in different states of the carrier and the corresponding actual monitoring data are obtained, and the data are preprocessed;

[0033] According to the thermal state parameter prediction data and the corresponding actual monitoring data, the rolling average error of the predicted value of each thermal state parameter is calculated respectively;

[0034] According to the rolling average error of the predicted value of each thermal state parameter, the consistency of each thermal state parameter is judged respectively, which is used for screening the reliability of the prediction data and the corresponding actual monitoring data, wherein if the rolling average error of a certain thermal state parameter exceeds the corresponding preset threshold for three times in succession, an alarm is triggered, and the parameter data will depend on the monitoring data rather than the prediction, and if not, the predicted parameter is considered reliable;

[0035] The rolling average error value expression of the two data is:

[0036] MAE_r j (k) = δ·MAE_r j (k-1) + (1-δ)·|Y Measurement,j (k) - Y pred.j (k|k-Δt)|

[0037] wherein, MAE_r j (k) is the rolling average error of the j-th thermal state parameter correction at the k-th calculation time, k is the index of the error calculation time, δ is a smoothing factor, MAE_r j (k-1) is the rolling average error of the j-th thermal state parameter correction at the k-1-th calculation time, Y Measurement，j (k) is the actual monitoring data of the j-th thermal state parameter at the k-th calculation time, Y pred，j(k|k-Δt) is the prediction data of the jth thermal state parameter at the kth calculation time, and Δt is the prediction time span of the model.

[0038] Preferably, the online gradient descent algorithm is combined with the driving data of the carrier to dynamically optimize the carrier general thermal state prediction base model, so that it automatically learns the driving habits, which specifically includes:

[0039] A fixed-size first-in-first-out (FIFO) buffer is set up to store the last N driving habit data pairs;

[0040] An incremental learning mechanism is set up to automatically trigger incremental learning of the carrier general thermal state prediction base model based on the rolling average error results of the data and the number of driving habit data pairs in the FIFO buffer;

[0041] Based on the incremental learning mechanism, the carrier general thermal state prediction base model is dynamically optimized to automatically learn the driving habits.

[0042] Preferably, the output results of the carrier general thermal state prediction base model are used to dynamically optimize the coolant flow rate based on the heat balance equation, which specifically includes:

[0043] The output results of the carrier general thermal state prediction base model are used as input values, and the optimized coolant flow rate is used as output values;

[0044] Based on the heat balance formula, a physical mapping model of the carrier thermal state parameters and the coolant flow rate is constructed, so that the parameter changes strictly follow the physical rules;

[0045] The output results of the carrier general thermal state prediction base model are input into the physical mapping model of the carrier thermal state parameters and the coolant flow rate to calculate the predicted value of the future battery pack self-accumulated heat power;

[0046] According to the top temperature control target, a target value of the battery pack self-accumulated heat power is dynamically generated;

[0047] In order to minimize the error between the battery pack self-accumulated heat power and the target value and to smooth the change of the coolant flow rate, a flow rate optimal control objective function under heat balance constraints is constructed;

[0048] In each control period, based on the gradient descent algorithm, the minimization problem of the objective function is solved to obtain the optimal coolant flow rate sequence for a future period of time;

[0049] The first control quantity in the optimal coolant flow rate sequence for a future period of time is output as the actual control instruction to the coolant pump actuator, and the above rolling optimization steps are repeated in the next period;

[0050] The expression of the physical mapping model of the carrier thermal state parameters and the coolant flow rate is:

[0051]

[0052] P = P + P 电池 P is the battery heat generation power, P 冷却液 P is the heat power taken away by the cooling liquid circulation, P 电池包 P is the heat power accumulated by the battery pack itself, c p Cp is the constant-pressure specific heat capacity of the cooling liquid, p is the density of the cooling liquid, v is the flow rate of the cooling liquid, T out T is the outlet temperature of the battery pack cooling liquid, T in T is the inlet temperature of the battery pack cooling liquid, m is the total mass of the battery pack, c 电池 Cp is the specific heat capacity of the battery, dT 电池 / dt is the change rate of the average temperature of the battery;

[0053] The flow rate optimal control objective function expression under the heat balance constraint is:

[0054]

[0055] min J = P + P 电池包，pred (k) is the predicted value of the heat power accumulated by the battery pack at the kth calculation time, P 电池包，set (k) is the target heat power accumulated by the battery pack at the kth calculation time, theta is a dimensionless equilibrium constant, which is adjusted through experiments or simulations to balance temperature accuracy and flow rate stability, v(k) is the predicted value of the cooling liquid flow rate at the kth calculation time, and v(k-1) is the predicted value of the cooling liquid flow rate at the (k-1)th calculation time.

[0056] Further, the present scheme proposes a heat pipe control parameter dynamic optimization system for a liquid-cooled lithium iron phosphate battery, which is used to realize the heat pipe control parameter dynamic optimization method for a liquid-cooled lithium iron phosphate battery as described above, comprising:

[0057] A data acquisition and database module is used to construct an offline training database of the carrier thermal state based on carrier state, thermal state, cooling system and environmental parameters, and to perform data cleaning and alignment; a data acquisition module is used to set a sensor group for real-time acquisition of current state parameters required for model prediction and actual thermal state data at future time for verifying prediction accuracy;

[0058] A thermal management optimization module is used for constructing a carrier general thermal state prediction base model based on a LightGBM model and a carrier thermal state offline training database, for predicting a carrier thermal state; a model verification unit is used for verifying whether the predicted carrier thermal state is reliable according to the carrier general thermal state prediction base model and actual monitoring results; an incremental learning unit is used for dynamically optimizing the carrier general thermal state prediction base model based on an online gradient descent algorithm and driving data of the carrier, so that the carrier general thermal state prediction base model automatically learns driving habits; and a thermal management optimization unit is used for dynamically optimizing a cooling liquid flow rate based on a heat balance equation and an output result of the carrier general thermal state prediction base model.

[0059] Preferably, the data acquisition and database module comprises:

[0060] A database unit is used for constructing a carrier thermal state offline training database based on carrier states, thermal states, cooling systems and environmental parameters, and performing data cleaning and alignment;

[0061] A data acquisition unit is used for setting a sensor group, for real-time acquisition of current state parameters required by the model prediction and thermal state actual data at future time points for verifying prediction accuracy.

[0062] Preferably, the thermal management optimization module comprises:

[0063] A model prediction unit is used for constructing a carrier general thermal state prediction base model based on a LightGBM model and a carrier thermal state offline training database, for predicting a carrier thermal state;

[0064] A model verification unit is used for verifying whether the predicted carrier thermal state is reliable according to the carrier general thermal state prediction base model and actual monitoring results;

[0065] An incremental learning unit is used for dynamically optimizing the carrier general thermal state prediction base model based on an online gradient descent algorithm and driving data of the carrier, so that the carrier general thermal state prediction base model automatically learns driving habits;

[0066] A thermal management optimization unit is used for dynamically optimizing a cooling liquid flow rate based on a heat balance equation and an output result of the carrier general thermal state prediction base model.

[0067] Compared with the prior art, the present application has the following beneficial effects:

[0068] The application provides a heat control parameter dynamic optimization method for a liquid-cooled lithium iron phosphate battery. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 The flow chart of the heat control parameter dynamic optimization method for the liquid-cooled lithium iron phosphate battery of the application;

[0070] Figure 2 The flow chart of the heat control parameter dynamic optimization method for the liquid-cooled lithium iron phosphate battery of the application;

[0071] Figure 3 The flow chart of the heat control parameter dynamic optimization method for the liquid-cooled lithium iron phosphate battery of the application;

[0072] Figure 4 The flow chart of the heat control parameter dynamic optimization method for the liquid-cooled lithium iron phosphate battery of the application; DETAILED DESCRIPTION

[0073] The following description is provided to enable any person skilled in the art to practice the application. The preferred embodiments described in the following description are only examples of the application.

[0074] REFERENCE Figure 1 As shown in the figure, a heat control parameter dynamic optimization method for a liquid-cooled lithium iron phosphate battery comprises:

[0075] Based on the state, thermal state, cooling system and environmental parameters of the carrier, a carrier thermal state offline training database is constructed, and data cleaning and alignment are performed;

[0076] According to the offline training database of the carrier thermal state, a carrier general thermal state prediction basic model is constructed based on a LightGBM model, which is used to predict the carrier thermal state;

[0077] According to the carrier general thermal state prediction basic model, combined with the actual monitoring results, it is verified whether the predicted carrier thermal state is reliable;

[0078] Based on the online gradient descent algorithm, combined with the driving data of the carrier, the carrier general thermal state prediction basic model is dynamically optimized, so that it can automatically learn the driving habits;

[0079] According to the output results of the carrier general thermal state prediction basic model, based on the heat balance equation, the cooling liquid flow rate is dynamically optimized;

[0080] Set up a sensor group to collect the current state parameters required by the model prediction and the actual thermal state data at future time for verifying the prediction accuracy in real time.

[0081] It can be explained that the thermal management performance of the liquid-cooled lithium iron phosphate battery is mainly affected by the cooling liquid type, the cell arrangement, the battery structure, the liquid cooling plate design, the heat conducting medium and the cooling liquid flow rate, etc. Among them, the cooling liquid type and the cell arrangement are inherent design and cannot be dynamically adjusted, while the cooling liquid flow rate and the carrier motion state (strongly related to driving habits) are key dynamic variables. The present application realizes the advanced prediction of the thermal state by fitting the mapping relationship between the carrier motion state and the thermal state, and then dynamically adjusts the cooling liquid flow rate to achieve the purpose of advanced thermal management. Therefore, the present scheme forms a complete dynamic optimization technical scheme through the whole process closed loop of offline database construction→ general model establishment→ prediction reliability verification→ online incremental learning→ flow rate optimization under physical constraints, thereby effectively solving the core defects of the prior art such as strong dependence on data volume, insufficient generalization ability, model "black box" and lack of physical constraints, and realizing the unification of high-precision prediction and personalized adaptive control under limited sample conditions.

[0082] The offline training database of the carrier thermal state is constructed based on the carrier state, the thermal state, the cooling system and the environmental parameters, and the data cleaning and alignment specifically includes:

[0083] Based on the carrier state parameters, the thermal state parameters, the cooling system parameters and the environmental parameters, the thermal management parameter types of the liquid-cooled lithium iron phosphate battery are determined;

[0084] Different carriers, battery capacities and cooling system architectures are selected to simulate the environment and the carrier motion to carry out test experiments, and the thermal management parameter samples of the liquid-cooled lithium iron phosphate battery are obtained;

[0085] Based on the median filtering algorithm, the collected parameter sample data is filtered to eliminate the influence of abnormal peak values;

[0086] Based on the min-max normalization formula, the collected parameter sample data is normalized to map the data to the interval [0, 1] and eliminate the influence of data dimension;

[0087] Based on the cubic spline interpolation method, the data is interpolated to ensure that the data of all channels is strictly aligned to the unified time axis;

[0088] An offline training database of carrier thermal state is constructed to store and record the thermal management parameter sample data of liquid-cooled lithium iron phosphate battery, facilitating subsequent parameter adjustment and update.

[0089] It can be explained that the carrier state parameters are: throttle pedal opening, brake pedal opening, motor request torque, motor actual power, carrier speed, longitudinal acceleration, battery current, battery total voltage, battery SOC; cooling system parameters are: cooling liquid actual flow rate, electronic water pump power, electronic water pump actual speed, radiator fan actual speed, fan power, PTC heater actual power, PTC heater outlet water temperature; environmental parameters are: ambient temperature, ambient humidity, atmospheric pressure; thermal state parameters are: battery average temperature, cooling liquid inlet temperature, cooling liquid outlet temperature; Since these parameters (such as throttle, power, current) directly determine the inflow and outflow of energy, and the generation and exchange of heat are the inevitable result of energy conversion, which follows the physical law (Joule's law, thermodynamic law), through the simulation of the covering environment cabin of the carrier, the chassis dynamometer standard cycle and the extreme test of multiple scene actual roads, the thermal management parameter samples under different working conditions are systematically collected, and the test content includes: high and low temperature environment adaptability evaluation, steady-state and transient load response, influence of different driving styles and road conditions on thermal behavior, etc. Key data such as battery heat generation, cooling system efficiency and temperature distribution are comprehensively obtained. Through data cleaning, alignment and fusion under these multi-dimensional and multi-boundary conditions, a high-quality and high-coverage offline training database is constructed to provide sufficient and reliable sample support for subsequent data-driven and physical model integrated thermal state prediction.

[0090] Referring to Figure 2 , the carrier general thermal state prediction base model is constructed based on the carrier thermal state offline training database and the LightGBM model, which is used to predict the carrier thermal state, specifically including:

[0091] The carrier state parameters, cooling system parameters and environmental parameters are taken as input quantities, and the carrier thermal state in the future period of time is taken as output quantity;

[0092] Among them, the environmental parameters are used to represent the dynamic constraints on the cooling liquid inlet temperature and the system heat dissipation boundary, so as to improve the prediction accuracy of the model under different working conditions;

[0093] Based on the carrier thermal state offline training database, it is divided into training set, validation set and test set according to the proportion of 7:2:1;

[0094] Based on the LightGBM framework, the built-in gradient boosting decision tree algorithm is used to construct a general thermal state prediction model for the carrier, which is composed of multiple regression trees.

[0095] Each regression tree splits nodes to minimize weighted squared error. When splitting, the optimal feature and its split threshold are selected by maximizing gain.

[0096] Set the initial learning rate A. During the training process, continuously monitor the loss on the validation set. If the loss does not decrease for 5 consecutive iterations, reduce the current learning rate by half

[0097] Based on the mean square error formula, a custom loss function with safety penalty is constructed to minimize the average temperature deviation of the battery.

[0098] Gradient boosting algorithm is used to optimize model parameters. In each iteration, the first and second order gradients of the current model with respect to the custom loss function are calculated, and a new regression tree is added to fit the negative gradient direction, thereby reducing the loss.

[0099] When the absolute value of the change of the validation set loss function value in K consecutive iterations is less than the preset threshold a, it is determined that the model has converged, and the training process is terminated in advance to avoid overfitting.

[0100] The gain expression is:

[0101]

[0102] In the formula, Gain is the gain value, which is used to evaluate the pros and cons of the split point, G L 、G R are the sum of the first order gradients of all samples in the left and right child nodes after splitting, H L 、H R are the sum of the second order gradients of all samples in the left and right child nodes after splitting, λ is the L2 regularization coefficient, and γ is the minimum gain threshold required for splitting. When Gain> γ, splitting will be performed.

[0103] The expression of the custom loss function with safety penalty is:

[0104]

[0105] In the formula, L is the custom loss value with safety penalty, M is the number of samples participating in training, Y true.iY is the real value of the average battery temperature of the i-th sample pred.i Y is the predicted value of the average battery temperature of the i-th sample safe α is the preset safety temperature threshold, and α is the safety penalty factor.

[0106] It can be explained that the LightGBM model as a lightweight gradient boosting machine learning framework can cope with the high-dimensional input of the battery thermal state prediction containing carrier state parameters, cooling system parameters and environmental parameters, etc. Its integrated algorithm based on decision tree is naturally good at processing tabular data and can automatically capture the complex nonlinear relationship and interaction effect between features without complex feature engineering or dimension transformation like neural networks. Secondly, its built-in gradient boosting decision tree algorithm can usually achieve higher precision than single models (such as SVR) or traditional neural networks under small samples by iteratively correcting errors, and effectively controls the model complexity through its built-in L2 regularization function to prevent overfitting and ensure generalization performance. Therefore, this scheme is based on the LightGBM framework, and uses its built-in gradient boosting decision tree algorithm to build a carrier general thermal state prediction base model to fit the mapping relationship between carrier state parameters, cooling system parameters, environmental parameters and carrier thermal state parameters, thereby overcoming the hysteresis of traditional feedback control based on the current state, realizing the advance prediction of the future temperature change of the battery, and facilitating the incremental learning and optimization of subsequent driving habit small sample data, and improving the personalization of the carrier general thermal state prediction base model. In this scheme, the environmental parameters (especially the environmental temperature) are used to represent the dynamic constraints of the cooling system heat dissipation boundary. Specifically, the environmental temperature directly limits the minimum theoretical outlet temperature of the radiator, thereby determining the lower limit of the cooling liquid inlet temperature. In a high-temperature environment, the system heat dissipation capacity decreases, and the cooling liquid's ability to carry heat weakens, causing the battery to heat up more easily. Therefore, the environmental parameters are crucial as model inputs, enabling the model to perceive changes in the external thermal environment and dynamically adjust its prediction benchmark, thereby significantly improving the prediction accuracy and generalization ability of the model under complex working conditions such as different regions and different seasons, and avoiding prediction errors that violate the principles of thermodynamics. Parameters λ and γ are hyperparameters that are determined by defining a search range, using cross-validation or evaluating the performance of the model under different λ values on the validation set, and selecting the λ value that performs best on the validation set. Parameter α is optimized based on the validation set by evaluating the model under different α values on the validation set, and selecting the α value that can significantly reduce the number of predicted temperatures exceeding the safety threshold Y safe without significantly worsening the overall prediction error. Parameter Y safe is a non-adjustable hard parameter determined by the properties of the battery itself, which can be obtained from the specification book of the battery cell manufacturer to query the working temperature range and irreversible damage temperature threshold of the battery cell.

[0107] The method according to the carrier universal thermal state prediction basic model, combined with the actual monitoring result, verifies whether the predicted carrier thermal state is reliable, specifically comprising:

[0108] According to the trained carrier universal thermal state prediction basic model, based on the actual monitoring of the carrier state parameter, the cooling system parameter and the environment parameter data, the carrier thermal state parameter data in the future period of time is obtained;

[0109] The thermal state parameter prediction data of the model under different states of the carrier and the corresponding actual monitoring data are obtained, and the data are preprocessed;

[0110] According to the thermal state parameter prediction data and the corresponding actual monitoring data, the rolling average error of the predicted value of each thermal state parameter is calculated;

[0111] According to the rolling average error of the predicted value of each thermal state parameter, the consistency of each thermal state parameter is judged respectively, which is used for screening the credibility of the prediction data and the corresponding actual monitoring data, wherein if the rolling average error of a certain thermal state parameter exceeds the corresponding preset threshold value for three times in succession, an alarm is triggered, and the parameter data will depend on the monitoring data rather than the prediction, and if not, it is considered that the predicted parameter is reliable;

[0112] The rolling average error value expression of the two data is:

[0113] MAE_r j (k)=δ·MAE_r j (k-1)+(1-δ)·|Y Measurement,j (k)-Y pred.j (k|k-Δt)|

[0114] In the formula, MAE_r j (k) is the rolling average error of the jth thermal state parameter correction at the kth calculation time, k is the index of the error calculation time, δ is a smoothing factor, MAE_r j (k-1) is the rolling average error of the jth thermal state parameter correction at the k-1th calculation time, Y Measurement，j (k) is the actual monitoring data of the jth thermal state parameter at the kth calculation time, Y pred，j (k|k-Δt) is the prediction data of the jth thermal state parameter at the kth calculation time, and Δt is the prediction time span of the model.

[0115] It can be explained that any data-driven model will gradually decline in performance after deployment due to changes in the external environment or encountering new conditions (such as extreme driving behavior) that were not seen during training. This phenomenon is called "model drift". By continuously comparing predicted values with true values, the decay trend of model performance can be monitored in real time. Once systematic prediction errors are found, an alarm is triggered, prompting the system to be maintained or retrained. It is dangerous to fully trust model predictions. If the model outputs an incorrect but seemingly reasonable prediction value for unknown reasons, such as predicting a lower temperature than the actual temperature, directly using it for control may cause the risk of battery overheating. Therefore, in the case where the prediction value cannot be guaranteed to be accurate, monitoring data rather than prediction can be relied on to reduce the risk of model misjudgment. Secondly, it provides a limiting condition for subsequent incremental learning. Incremental learning is not performed at all times. It needs to be triggered at the right time to avoid unnecessary calculations or learning noise. The rolling average error is a key indicator for determining whether the model needs to learn. When the rolling average error is higher than its historical baseline value for several times in a row but does not reach the failure threshold, it is used as a signal to trigger the incremental learning mechanism in the subsequent steps. New driving data is used to fine-tune the model to adapt to changes. The smoothing factor δ is determined by grid search on the validation set, and its typical empirical value is 0.9.

[0116] Referring to Figure 3 The online gradient descent algorithm is combined with the driving data of the carrier to dynamically optimize the carrier universal thermal state prediction base model to automatically learn driving habits, which specifically includes:

[0117] A fixed-size first-in-first-out (FIFO) buffer is set up to store the last N driving habit data pairs;

[0118] An incremental learning mechanism is set up to automatically trigger incremental learning of the carrier universal thermal state prediction base model according to the rolling average error of the data and the number of driving habit data pairs in the FIFO buffer;

[0119] Based on the incremental learning mechanism, the carrier universal thermal state prediction base model is dynamically optimized to automatically learn driving habits.

[0120] It can be explained that, based on the carrier universal thermal state prediction base model, the present scheme learns the driving habits of the carrier by introducing an incremental learning mechanism and combining driving habit data, so that the more sample data used to participate in the incremental learning mechanism, the more personalized and intelligent the dynamic optimization of the thermal management parameters of the carrier liquid-cooled lithium iron phosphate battery is. The driving habit data pair is a tuple [X(k), Y Measurement (k)], where X(k) is the carrier state input data at the kth calculation time, and Y Measurement(k) is the actual monitoring data of the thermal state at the kth calculation time, which is continuously collected to characterize driving habits and the systematic thermal response rules caused by them, such as the frequency of rapid acceleration and deceleration, the commonly used speed interval, and the typical thermal load;

[0121] The incremental learning mechanism includes an incremental learning triggering mechanism and an incremental learning updating mechanism. The incremental learning triggering mechanism refers to automatically triggering incremental learning when two conditions are met: the data volume in the FIFO buffer exceeds the preset number threshold, and the rolling average error is higher than the historical baseline value for multiple times but does not reach the failure threshold. The preset number threshold, historical baseline value, and failure threshold are determined by big data based on normal distribution and combined with expert experience.

[0122] The incremental learning updating mechanism includes:

[0123] All existing decision tree parameters in the trained carrier universal thermal state prediction base model are fixed as a fixed feature extractor, and a linear fully connected layer is added after the output (i.e., the sum of all decision tree prediction values).

[0124] A batch of driving habit data pairs is randomly sampled from the FIFO buffer to calculate the loss value of the carrier universal thermal state prediction base model on the current batch.

[0125] An online gradient descent algorithm is used to calculate the gradient of the loss function with respect to the new linear layer weights and biases. The error is backpropagated to update only the weights and biases in the new linear layer.

[0126] The updated parameters replace the original parameters in the carrier universal thermal state prediction base model for subsequent prediction.

[0127] A batch of driving habit data pairs is randomly sampled from the FIFO buffer to calculate the loss value of the carrier universal thermal state prediction base model on the current batch. A custom loss function with safety penalty is directly applied to ensure consistency between online incremental learning and offline training optimization objectives.

[0128] Referring to Figure 4 The output of the carrier universal thermal state prediction base model is used as the input value, and the cooling liquid flow rate optimization value is used as the output value.

[0129] The output of the carrier universal thermal state prediction base model is used as the input value, and the cooling liquid flow rate optimization value is used as the output value.

[0130] Based on the heat balance formula, a physical mapping model of carrier thermal state parameters and cooling liquid flow rate is constructed to strictly follow the physical rules of parameter changes.

[0131] The output result of the carrier general thermal state prediction base model is input into a carrier thermal state parameter and coolant flow rate physical mapping model to calculate a future battery pack self-accumulated heat power prediction value;

[0132] According to the top layer temperature control target, a target value of the battery pack self-accumulated heat power is dynamically generated;

[0133] A flow rate optimal control target function under a heat balance constraint is constructed for the purpose of minimizing the error between the battery pack self-accumulated heat power and the target value and smoothing the change of the coolant flow rate;

[0134] In each control period, a minimum problem of the target function is solved based on a gradient descent algorithm to obtain an optimal coolant flow rate sequence in a future period of time;

[0135] The first control amount in the optimal coolant flow rate sequence in the future period of time is taken as an actual control instruction output to a coolant pump actuator, and the above rolling optimization steps are repeated in the next period;

[0136] The expression of the carrier thermal state parameter and coolant flow rate physical mapping model is:

[0137]

[0138] In the formula, P 电池 is a battery heat generation power, P 冷却液 is a heat power taken away by a coolant circulation, P 电池包 is a battery pack self-accumulated heat power, c p is a constant-pressure specific heat capacity of the coolant, p is a coolant density, v is a coolant flow rate, T out is a battery pack coolant outlet temperature, T in is a battery pack coolant inlet temperature, m is a total mass of the battery pack, c 电池 is a battery specific heat capacity, dT 电池 / dt is a change rate of the average temperature of the battery;

[0139] The expression of the flow rate optimal control target function under the heat balance constraint is:

[0140]

[0141] In the formula, minJ is a flow rate optimal control target function value under the heat balance constraint, P 电池包，pred (k) is a battery pack self-accumulated heat power prediction value at the kth calculation time, P 电池包，set(k) is the heat power accumulated by the target battery pack itself at the kth calculation time, θ is a balance constant, dimensionless, which is adjusted by experiment or simulation, and the temperature accuracy and flow stability are balanced, v(k) is the predicted value of the cooling liquid flow rate at the kth calculation time, and v(k-1) is the predicted value of the cooling liquid flow rate at the (k-1)th calculation time.

[0142] It can be explained that the change of battery temperature is the result of heat power accumulation, even if perfect cooling is applied immediately, the temperature will not drop instantly, but will change slowly according to the physical law of heat capacity, if the temperature error is taken as the target, the controller is always chasing an already occurred, lagging fact, and the control response has a lag, while the heat power accumulated by the battery pack itself directly determines the future temperature change direction and rate, if the heat power accumulated by the battery pack itself is greater than 0, it means that the temperature rises, if the heat power accumulated by the battery pack itself is less than 0, it means that the temperature drops, by taking the heat power accumulated by the battery pack itself as the target, the heat power accumulated by the battery pack itself at this moment is controlled, which is equivalent to controlling the temperature trajectory in the future, so as to realize the real advance control of temperature, secondly, the heat power accumulated by the battery pack itself has a clear physical mapping model, through the physical mapping model, the optimal cooling liquid flow rate sequence in the future can be effectively limited without exceeding the physical limit, so as to be more accurate, wherein, P 电池包，set It is not a fixed value, but a set value dynamically generated according to the top temperature target, and its calculation formula is:

[0143] P 电池包，set = K p ·(T set -T pred (k))

[0144] The control meaning of the formula is: by adjusting the target heat power accumulated by the battery pack itself, the predicted value of the average temperature of the battery pack is driven to converge to the set value T set , wherein K p is a proportional coefficient, which can be preliminarily set according to the heat capacity mc 电池 of the battery pack.

[0145] The setting sensor group is used to collect the current state parameters required by the model prediction and the actual thermal state data at the future time for verifying the prediction accuracy, which specifically includes:

[0146] The sensor group is disposed at the key positions of the carrier, used to collect the thermal management parameter data of the liquid-cooled lithium iron phosphate battery, and the output data of the sensor group is divided into two types of flow directions.

[0147] The sensor group includes:

[0148] Temperature sensor array: used to monitor thermal state parameters, deployment locations include battery pack internal cell surface, cooling liquid inlet pipeline, cooling liquid outlet pipeline;

[0149] Carrier state sensor: used to monitor carrier motion parameters, obtained through carrier CAN bus;

[0150] Cooling system sensor: used to monitor actuator state, including cooling liquid flow rate sensor, electronic water pump rotation speed sensor;

[0151] The output data of the sensor group is divided into two categories and flows to the specific:

[0152] The carrier state parameters, cooling system parameters and environmental parameters are input in real time to the carrier general thermal state prediction basic model for driving the model to predict future thermal state;

[0153] The actual monitoring data of the thermal state parameters are compared with the prediction results of the model to calculate the rolling average error and verify the reliability of the prediction results to provide feedback for optimization control;

[0154] The hardware synchronous acquisition method is adopted to ensure that all sensor data have unified and synchronous time stamps and guarantee the consistency of the data in the time dimension.

[0155] It can be explained that the sensor group is set to collect liquid-cooled lithium iron phosphate battery thermal control parameter data in real time, which aims to provide input and verification data for the carrier general thermal state prediction basic model to realize incremental optimization of the model, serve as a safety redundant guarantee system for degradation control to prevent thermal runaway when the model fails, provide necessary input support for the physical mapping model to optimize control of thermal balance, and directly serve the calculation of the optimal cooling liquid flow rate in the real-time control cycle. Full-dimensional guarantee of the implementability, safety and real-time optimization execution of thermal control.

[0156] Further, based on the same inventive concept as the above-mentioned dynamic optimization method for thermal control parameters of liquid-cooled lithium iron phosphate batteries, the present scheme proposes a dynamic optimization system for thermal control parameters of liquid-cooled lithium iron phosphate batteries, comprising:

[0157] The data acquisition and database module is used to construct the carrier thermal state offline training database based on the carrier state, thermal state, cooling system and environmental parameters, and to perform data cleaning and alignment; the data acquisition module is used to set the sensor group to collect the current state parameters required by the model prediction and the thermal state actual data at the future time for verifying the prediction accuracy in real time;

[0158] The thermal management optimization module is used for constructing a carrier general thermal state prediction base model based on a LightGBM model and a carrier thermal state offline training database, and is used for predicting the carrier thermal state; the model verification unit is used for verifying whether the predicted carrier thermal state is reliable according to the carrier general thermal state prediction base model and actual monitoring results; the incremental learning unit is used for dynamically optimizing the carrier general thermal state prediction base model based on an online gradient descent algorithm and driving data of the carrier, so that the carrier general thermal state prediction base model automatically learns driving habits; and the thermal management optimization unit is used for dynamically optimizing a cooling liquid flow rate based on a heat balance equation according to an output result of the carrier general thermal state prediction base model.

[0159] The data acquisition and database module comprises:

[0160] The database unit is used for constructing a carrier thermal state offline training database based on carrier states, thermal states, cooling systems and environmental parameters, and performing data cleaning and alignment;

[0161] The data acquisition unit is used for setting a sensor group, and is used for acquiring current state parameters required by the model prediction and thermal state actual data for verifying prediction accuracy at a future time in real time;

[0162] The thermal management optimization module comprises:

[0163] The model prediction unit is used for constructing a carrier general thermal state prediction base model based on a LightGBM model and a carrier thermal state offline training database, and is used for predicting the carrier thermal state;

[0164] The model verification unit is used for verifying whether the predicted carrier thermal state is reliable according to the carrier general thermal state prediction base model and actual monitoring results;

[0165] The incremental learning unit is used for dynamically optimizing the carrier general thermal state prediction base model based on an online gradient descent algorithm and driving data of the carrier, so that the carrier general thermal state prediction base model automatically learns driving habits;

[0166] The thermal management optimization unit is used for dynamically optimizing a cooling liquid flow rate based on a heat balance equation according to an output result of the carrier general thermal state prediction base model.

[0167] In summary, the application has the advantages that by fusing data driving and a physical model, online learning is used to dynamically optimize thermal management parameters, and the thermal management precision, adaptability and safety of the liquid-cooled lithium iron phosphate battery are significantly improved.

[0168] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic optimization of thermal management parameters for liquid-cooled lithium iron phosphate batteries, characterized in that, The application comprises the following steps: Based on the carrier state, thermal state, cooling system and environmental parameters, a carrier thermal state offline training database is constructed, and data cleaning and alignment are performed; According to the carrier thermal state offline training database, a carrier general thermal state prediction base model is constructed based on the LightGBM model, which is used to predict the carrier thermal state; According to the carrier general thermal state prediction base model, combined with the actual monitoring results, whether the predicted carrier thermal state is reliable is verified; Based on the online gradient descent algorithm, combined with the driving data of the carrier, the carrier general thermal state prediction base model is dynamically optimized to automatically learn the driving habits; According to the output results of the carrier general thermal state prediction base model, based on the heat balance equation, the cooling liquid flow rate is dynamically optimized; A sensor group is set up to collect the current state parameters required by the model prediction and the actual thermal state data for verifying the prediction accuracy at future time.

2. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 1, characterized in that, The construction of the carrier thermal state offline training database based on the carrier state, thermal state, cooling system and environmental parameters, and the data cleaning and alignment specifically comprise the following steps: Based on the carrier state parameters, thermal state parameters, cooling system parameters and environmental parameters, the thermal management parameter types of the liquid-cooled lithium iron phosphate battery are determined; Different carriers, battery capacities and cooling system architectures are selected to simulate the environment and carrier movement to carry out test experiments, and the thermal management parameter samples of the liquid-cooled lithium iron phosphate battery are obtained; Based on the median filter algorithm, the collected parameter sample data is filtered to eliminate the influence of abnormal peak values; Based on the min-max normalization formula, the collected parameter sample data is normalized to map the data to the [0, 1] interval, eliminating the influence of data dimension; Based on the cubic spline interpolation method, the data is interpolated to ensure that the data of all channels is strictly aligned to a unified time axis; The carrier thermal state offline training database is constructed to store and record the thermal management parameter sample data of the liquid-cooled lithium iron phosphate battery, facilitating subsequent parameter adjustment and update.

3. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 2, characterized in that, The construction of the carrier general thermal state prediction base model based on the carrier thermal state offline training database and the LightGBM model for predicting the carrier thermal state specifically comprises the following steps: The carrier state parameters, cooling system parameters and environmental parameters are taken as input, and the carrier thermal state parameters in the future period of time are taken as output; Among them, the environmental parameters are used to represent the dynamic constraints of the cooling liquid inlet temperature and the system heat dissipation boundary, so as to improve the prediction accuracy of the model under different working conditions; Based on the carrier thermal state offline training database, the training set, the validation set and the test set are divided in the ratio of 7:2:1; Based on the LightGBM framework, the built-in gradient boosting decision tree algorithm is used to construct the carrier general thermal state prediction base model, wherein the model is composed of multiple regression trees; Each regression tree splits the nodes with the minimum weighted squared error minimization as the target, and selects the optimal feature and its split threshold by maximizing the gain (Gain) during splitting; Set the initial learning rate as A, during the training process, continuously monitor the loss on the validation set, if the loss does not decrease for 5 consecutive iterations, then halve the current learning rate Based on the mean square error formula, a self-defined loss function with safety penalty is constructed to minimize the average temperature deviation of the battery. The gradient boosting algorithm is used to optimize the model parameters, that is, in each iteration, the first-order gradient and second-order gradient of the current model with respect to the custom loss function are calculated, a new regression tree is added to fit the negative gradient direction, thereby reducing the loss; When the absolute value of the change of the loss function value of the verification set in the last K iterations is less than a preset threshold a, it is determined that the model has converged, and the training process is terminated in advance to avoid overfitting; The gain expression is: In the formula, Gain is a gain value used to evaluate the pros and cons of the split point, G L , G R are the sums of the first-order gradients of all samples in the left and right child nodes after the split, H L , H R are the sums of the second-order gradients of all samples in the left and right child nodes after the split, λ is an L2 regularization coefficient, and γ is a minimum gain threshold required for the split, wherein, Splitting is performed only when Gain>γ. The expression of the custom loss function with safety penalty is: In the formula, L is a user-defined loss value with a safety penalty, M is the number of samples participating in training, Y true.i is the real value of the average battery temperature of the i th sample, Y pred.i is the predicted value of the average battery temperature of the i th sample, Y safe is a preset safety temperature threshold, and a is a safety penalty factor.

4. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 3, characterized in that, The specific steps of verifying whether the predicted carrier thermal state is reliable according to the carrier universal thermal state prediction base model combined with the actual monitoring result include: According to the trained carrier universal thermal state prediction base model, based on the actual monitoring data of the carrier state parameters, cooling system parameters and environmental parameters, the carrier thermal state parameter data in the future period of time is obtained; The thermal state parameter prediction data of the model under different states of the carrier and the corresponding actual monitoring data are obtained, and the data are preprocessed; According to the thermal state parameter prediction data and the corresponding actual monitoring data, the rolling average error of the predicted value of each thermal state parameter is calculated; According to the rolling average error of the predicted value of each thermal state parameter, the consistency of each thermal state parameter is judged respectively, which is used to filter the reliability of the prediction data and the corresponding actual monitoring data, wherein if the rolling average error of a certain thermal state parameter exceeds the corresponding preset threshold for three times in a row, an alarm is triggered, and the parameter data will depend on the monitoring data rather than the prediction, and if not, the predicted parameter is considered reliable; The expression of the rolling average error value of the two data is: MAE_r j (k) = δ · MAE_r j (k-1) + (1 - δ) · |Y Measurement,j (k) - Y pred.j (k|k-Δt) | wherein MAE_r j (k) is the rolling average error of the jth thermal state parameter correction at the kth calculation time, k is the index of the error calculation time, δ is the smoothing factor, MAE_r j (k-1) is the rolling average error of the jth thermal state parameter correction at the k-1th calculation time, Y Measurement，j (k) is the actual monitoring data of the jth thermal state parameter at the kth calculation time, Y pred，j (k|k-Δt) is the predicted data of the jth thermal state parameter at the kth calculation time, Δt is the prediction time span of the model.

5. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 4, characterized in that, The specific steps of dynamically optimizing the carrier universal thermal state prediction base model based on the online gradient descent algorithm combined with the driving data of the carrier to make it automatically learn the driving habits include: A fixed-size first-in-first-out (FIFO) buffer is set up to store the last N pairs of driving habit data; An incremental learning mechanism is set up to automatically trigger the incremental learning of the carrier universal thermal state prediction base model according to the rolling average error result of the data and the number of pairs of driving habit data in the FIFO buffer; Based on the incremental learning mechanism, the carrier universal thermal state prediction base model is dynamically optimized to automatically learn the driving habits.

6. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 5, characterized in that, The specific steps of dynamically optimizing the cooling liquid flow rate based on the heat balance equation according to the output result of the carrier universal thermal state prediction base model include: The output result of the carrier universal thermal state prediction base model is taken as the input value, and the optimized value of the cooling liquid flow rate is taken as the output value; Based on the heat balance formula, a physical mapping model of the carrier thermal state parameters and the cooling liquid flow rate is constructed, so that the parameter changes strictly follow the physical rules; The output result of the carrier universal thermal state prediction base model is input into the physical mapping model of the carrier thermal state parameters and the cooling liquid flow rate to calculate the predicted value of the heat power accumulated by the battery pack itself in the future; According to the top temperature control target, the target value of the heat power accumulated by the battery pack itself is dynamically generated; In order to minimize the error between the heat power accumulated by the battery pack itself and the target value and to smooth the change of the cooling liquid flow rate, a flow rate optimal control objective function under the heat balance constraint is constructed; In each control cycle, based on the gradient descent algorithm, the minimization problem of the objective function is solved to obtain the optimal coolant flow rate sequence in the future period of time; The first control quantity in the optimal coolant flow rate sequence in the future period of time is taken as the actual control instruction and output to the coolant pump actuator, and the above rolling optimization step is repeated in the next cycle; The physical mapping model expression of the carrier thermal state parameter and the coolant flow rate is: where P 电池 is the battery heat generation power, P 冷却液 is the heat power taken away by the coolant circulation, P 电池包 is the heat power accumulated by the battery pack itself, c p is the specific heat capacity of the coolant at constant pressure, p is the density of the coolant, v is the flow rate of the coolant, T out is the outlet temperature of the battery pack coolant, T in is the inlet temperature of the battery pack coolant, m is the total mass of the battery pack, c 电池 is the specific heat capacity of the battery, dT 电池 / dt is the rate of change of the average temperature of the battery. The flow rate optimal control objective function expression under the thermal balance constraint is: where minJ is the optimal control objective function value of flow rate under the constraint of thermal equilibrium, P 电池包，pred (k) is the predicted value of the heat power accumulated by the battery pack at the kth calculation time, P 电池包，set (k) is the heat power accumulated by the target battery pack at the kth calculation time, θ is the equilibrium constant, dimensionless, and is adjusted through experiments or simulations to balance the temperature accuracy and flow rate stability, v(k) is the predicted value of the coolant flow rate at the kth calculation time, and v(k-1) is the predicted value of the coolant flow rate at the (k-1)th calculation time.

7. The method for dynamic optimization of thermal management parameters for liquid-cooled LiFePO4 battery according to claim 6, characterized in that, The sensor group is arranged for real-time acquisition of the current state parameter required for model prediction and the actual thermal state data at the future time for verifying the prediction accuracy, and specifically includes: The sensor group is arranged for real-time acquisition of the current state parameter required for model prediction and the actual thermal state data at the future time for verifying the prediction accuracy, and specifically includes: The sensor group includes: A temperature sensor array is arranged for monitoring the thermal state parameter, and the deployment position includes the surface of the battery cell inside the battery pack, the coolant inlet pipeline and the coolant outlet pipeline; A carrier state sensor is arranged for monitoring the carrier motion parameter, and the carrier CAN bus is used for acquisition; A cooling system sensor is arranged for monitoring the actuator state, including a coolant flow rate sensor and an electronic water pump speed sensor; The output data of the sensor group is divided into two categories, and specifically includes: The carrier state parameter, the cooling system parameter and the environmental parameter are input to the carrier general thermal state prediction basic model in real time, for driving the model to predict the future thermal state; The actual monitoring data of the thermal state parameter is compared with the prediction result of the model, for calculating the rolling average error and verifying the reliability of the prediction result, and providing feedback for the optimal control; A hardware synchronous acquisition mode is adopted to ensure that all sensor data have unified and synchronous time stamps, and the consistency of the data in the time dimension is ensured.

8. A heat management parameter dynamic optimization system for liquid-cooled lithium iron phosphate batteries, characterized in that, The method comprises the following steps: A data acquisition and database module is arranged for constructing a carrier thermal state offline training database based on the carrier state, the thermal state, the cooling system and the environmental parameter, and performing data cleaning and alignment; a data acquisition module is arranged for arranging a sensor group for real-time acquisition of the current state parameter required for model prediction and the actual thermal state data at the future time for verifying the prediction accuracy; A thermal management optimization module is arranged for constructing a carrier general thermal state prediction basic model based on a LightGBM model, for predicting the carrier thermal state; verifying whether the predicted carrier thermal state is reliable according to the output result of the carrier general thermal state prediction basic model and the actual monitoring result; dynamically optimizing the carrier general thermal state prediction basic model based on an online gradient descent algorithm and the driving data of the carrier, so that the model automatically learns the driving habits; and dynamically optimizing the coolant flow rate based on the thermal balance equation and the output result of the carrier general thermal state prediction basic model.

9. The thermal management parameter dynamic optimization system for liquid-cooled lithium iron phosphate battery of claim 8, wherein, The data acquisition and database module includes: A database unit is configured to construct a carrier thermal state offline training database based on carrier states, thermal states, cooling systems and environmental parameters, and to perform data cleaning and alignment. A data acquisition unit is configured to set a sensor group to acquire current state parameters required for model prediction and thermal state actual data at future time points for verifying prediction accuracy in real time.

10. The thermal management parameter dynamic optimization system for liquid-cooled lithium iron phosphate battery of claim 9, wherein, The thermal management optimization module comprises: A model prediction unit is configured to construct a carrier general thermal state prediction base model based on a LightGBM model for predicting carrier thermal states based on the carrier thermal state offline training database. A model verification unit is configured to verify whether the predicted carrier thermal state is reliable based on the carrier general thermal state prediction base model and actual monitoring results. An incremental learning unit is configured to dynamically optimize the carrier general thermal state prediction base model based on an online gradient descent algorithm and driving data of the carrier to enable the model to automatically learn driving habits. A thermal management optimization unit is configured to dynamically optimize the cooling liquid flow rate based on a thermal balance equation based on the output results of the carrier general thermal state prediction base model.

Images